應用於H.264上的低複雜度提昇取樣技術

Abstract

高解析度的視訊規格不斷的隨著科技進步而提昇，對應所需的硬體規格也隨之提昇，使用者需要更精準的影像擷取感測器與更大的儲存記憶體來滿足要求，這是十分昂貴的代價，所以可使用在各式硬體平台上，將低解析度影像轉換成高解析度影像，或是將較小的影像提昇取樣至較大的影像的「超解析度重建技術」，成為許多應用的替代方案，至今仍然不斷的被研發與改進。 現今已逐漸被廣泛使用的視訊壓縮主流格為ISO與ITU-T合作在2003年制定的H.264/AVC，不同於傳統固定區塊的視訊壓縮編碼，其變動區塊的動作預測、動作補償與四分之一精度運動向量等特性，大幅減少了編碼冗餘，提昇視訊壓縮率。本文提出的可變動區塊超解析度重建技術是針對H.264/AVC中變動區塊所保留的影像相依性資訊，根據區塊的影像內容選擇對應的超解析度重建技術來降低整體的運算複雜度，加快提昇取樣(增加取樣)的速度。我們以區塊大小與預估模式將區塊分為關鍵區塊、複雜動作區塊以及未變動區塊等部份，只針對人眼視覺較感興趣的部份選擇提昇影像品質較多的演算法，以減少超解析度重建過程的時間，並使重建後的結果在整體上依然保有不錯的視覺品質。

High resolution image/video techniques are developed rapidly, it means they need more accuracy sensors to acquire images/videos and more capacity to store the data. They are usually expensive and not everyone can afford them. So the technique which can enhance the low resolution image(s) to higher one called “super-resolution” is quite important and many researches and algorithms were proposed in recent decades. Now, the popular video specification H.264/AVC which developed by Joint Video Team (JVT, organized by ISO and ITU-T) at 2003 was used to many applications. H.264/AVC is not like the traditional block-based video CODEC using fixed block size, it use the variable block size to make motion prediction and compensation more accuracy to reduce the coding redundancy and enhance compression rate. In this paper, we propose an approach “Variable block size super resolution” which focus the variable block size and predicted mode of H.264/AVC can base on different content into block to choose corresponding super-resolution algorithms. It can decrease whole complexity and make the up-sampling more efficient. This work first categorizes each block into several types: key blocks, complex motion vector blocks and constant blocks, then enhances more quality for the areas are interesting by human vision. This content adaptive strategy can reduce computational time and preserve acceptable visual quality. Experimental results are given to the efficiency of the proposed approach.